pith. sign in

arxiv: 2605.31302 · v1 · pith:3UJJJSLTnew · submitted 2026-05-29 · 📡 eess.IV · cs.CV· eess.SP

MoE-dqINR: A Unified Mixture-of-Experts Implicit Neural Representation Framework for Scan-Specific Dynamic and Quantitative MRI Reconstruction

Yinzhe Wu , Fanwen Wang , Zhenxuan Zhang , Zi Wang , Chengyan Wang , Guang Yang This is my paper

Pith reviewed 2026-06-28 20:07 UTC · model grok-4.3

classification 📡 eess.IV cs.CVeess.SP
keywords MRI reconstructionimplicit neural representationsmixture of expertsdynamic MRIquantitative MRIscan-specific reconstructionmulticoil k-space
0
0 comments X

The pith

State-conditioned mixture-of-experts INR unifies shared spatial representation with dynamic and quantitative MRI synthesis from undersampled multicoil data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces MoE-dqINR as a scan-specific multicoil reconstruction method that factorizes image representation into reusable spatial experts and a routing pathway driven by acquisition state. Spatial experts capture coordinate-dependent content that can be shared across frames or contrasts, while routing weights conditioned on the normalized state index select and combine experts to produce each state-specific image. This separation supports both dynamic and quantitative MRI without explicit motion models or sequence-specific signal equations. The approach couples the representation directly to the multicoil forward model and achieves per-scan optimization in roughly 30 seconds.

Core claim

The proposed formulation establishes state-conditioned mixture-of-experts INR as a scan-specific multicoil MRI reconstruction prior that unifies shared spatial representation, dynamic- and qMRI-specific synthesis, and practical per-scan efficiency.

What carries the argument

State-conditioned routing pathway that uses the normalized state index to compute weights over a bank of shared spatial experts, synthesizing each acquisition state from the common expert set.

If this is right

  • Reduces per-scan INR optimization time to approximately 30 seconds while preserving state-dependent fidelity.
  • Enables a single image-first architecture to handle both dynamic MRI and quantitative MRI without separate motion or deformation modules.
  • Allows spatial information to be reused across acquisition states through the shared expert bank.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The routing design could be tested on modalities beyond MRI that involve ordered state changes, such as dynamic CT or ultrasound series.
  • If the expert bank generalizes across patients, the per-scan training cost might drop further by initializing from a population-level expert set.
  • Direct comparison of expert activation patterns across states might reveal whether the method implicitly captures contrast evolution without explicit physics models.

Load-bearing premise

Conditioning routing weights solely on the normalized state index is sufficient to synthesize each dynamic frame or contrast state from a shared expert bank without loss of fidelity or the need for additional sequence-specific quantitative signal models.

What would settle it

Reconstruction error or artifact levels that remain high on a held-out dynamic or quantitative dataset when the model is trained only on the shared experts and state-index routing without retraining or extra signal equations.

Figures

Figures reproduced from arXiv: 2605.31302 by Chengyan Wang, Fanwen Wang, Guang Yang, Yinzhe Wu, Zhenxuan Zhang, Zi Wang.

Figure 1
Figure 1. Figure 1: Overview of MoE-dqINR. (A) Scan-specific MRI INR design space. Existing [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Main MoE-dqINR representation for dynamic and quantitative MRI reconstruction. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Representative cine reconstruction comparison (Cardiac region-of-interest (ROI)) [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Representative T1-mapping qualitative reconstruction comparison at (A, B) AF= [PITH_FULL_IMAGE:figures/full_fig_p018_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Top: State-conditioned expert routing behavior in representative cine and T1- mapping reconstructions at AF=6×. (A) For cine, the router combines K = 7 shared spatial experts across 12 temporal frames; (B) for T1 mapping, it combines K = 10 shared spatial experts across 9 ordered contrast states. Bottom: The corresponding routing matrices show the simplex-normalized expert weights αqk, with rows indexing e… view at source ↗
read the original abstract

Undersampled magnetic resonance imaging (MRI) reconstruction seeks to recover temporally or contrast-varying image series from incomplete multicoil k-space data while preserving state-dependent fidelity for dynamic and quantitative MRI (qMRI). Existing scan-specific implicit neural representations (INRs) often use monolithic spatiotemporal coordinate fields, explicit subspaces, motion or deformation models, calibration variables, or sequence-specific quantitative signal models. These design choices can limit flexibility in sharing spatial information while adapting image synthesis across acquisition states. Moreover, many INR-based baselines remain computationally demanding, typically requiring per-scan optimization times on the order of hundreds to thousands of seconds. We propose MoE-dqINR, a scan-specific multicoil MRI reconstruction framework that factorizes the image-domain representation into shared spatial experts and a state-conditioned routing pathway. Spatial experts encode reusable coordinate-dependent image content, whereas routing weights, conditioned on ordered acquisition states, synthesize each dynamic frame or contrast state from a common expert bank. The representation is coupled to a multicoil MRI forward model, uses the normalized state index to drive routing in both dynamic and quantitative MRI. By separating shared spatial representation from state-dependent synthesis, the framework provides an image-first architecture for dynamic and quantitative MRI while reducing scan-specific INR optimization to approximately 30 s per scan in our experiments. The proposed formulation establishes state-conditioned mixture-of-experts INR as a scan-specific multicoil MRI reconstruction prior that unifies shared spatial representation, dynamic- and qMRI-specific synthesis, and practical per-scan efficiency.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes MoE-dqINR, a scan-specific multicoil MRI reconstruction framework based on a mixture-of-experts implicit neural representation. It factorizes the representation into shared spatial experts encoding coordinate-dependent content and a state-conditioned routing pathway driven by the normalized acquisition state index to synthesize dynamic frames or contrast states from a common expert bank. The model is coupled only to the multicoil forward model and is claimed to unify dynamic and quantitative MRI while achieving approximately 30 s per-scan optimization.

Significance. If the central claims hold with supporting validation, the factorization into shared spatial experts plus scalar state routing could provide a flexible image-first prior that improves efficiency and adaptability over monolithic INRs or sequence-specific models for both dynamic and qMRI regimes.

major comments (2)
  1. [Abstract] Abstract (paragraph on routing pathway): the claim that conditioning routing weights solely on the normalized state index suffices to synthesize quantitative contrast states at full fidelity without sequence-specific quantitative signal models is load-bearing for the unification claim, yet no derivation, signal-physics justification, or empirical test is supplied to show that ordered acquisition index is an adequate proxy for nonlinear evolution (relaxation, flip-angle dependence, etc.).
  2. [Abstract] Abstract: the stated 30 s per-scan timing and unification benefit are presented without any quantitative results, error metrics, ablation studies, or baseline comparisons, so the central performance and unification assertions cannot be evaluated from the provided text.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'in our experiments' for the timing claim is used without reference to datasets, acquisition protocols, or hardware, reducing clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed feedback on the abstract. We address each major comment below and propose revisions where appropriate to strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract] Abstract (paragraph on routing pathway): the claim that conditioning routing weights solely on the normalized state index suffices to synthesize quantitative contrast states at full fidelity without sequence-specific quantitative signal models is load-bearing for the unification claim, yet no derivation, signal-physics justification, or empirical test is supplied to show that ordered acquisition index is an adequate proxy for nonlinear evolution (relaxation, flip-angle dependence, etc.).

    Authors: The manuscript presents the normalized state index as an empirical design choice for routing that enables unification across dynamic and quantitative regimes without explicit sequence-specific signal models. The full text includes supporting experiments on both task types that validate reconstruction fidelity. We agree the abstract would be strengthened by briefly noting this empirical basis and will revise it to reference the experimental validation while directing readers to the methods for the routing formulation. revision: yes

  2. Referee: [Abstract] Abstract: the stated 30 s per-scan timing and unification benefit are presented without any quantitative results, error metrics, ablation studies, or baseline comparisons, so the central performance and unification assertions cannot be evaluated from the provided text.

    Authors: We agree that the abstract would benefit from including representative quantitative support. The full manuscript reports specific timing, error metrics, and comparisons in the experiments. We will revise the abstract to incorporate key results (e.g., per-scan timing and performance metrics relative to baselines) to make the claims evaluable from the abstract alone. revision: yes

Circularity Check

0 steps flagged

No circularity detected; derivation is self-contained architectural proposal

full rationale

The paper presents MoE-dqINR as a new scan-specific INR architecture that factorizes shared spatial experts from state-conditioned routing weights driven by normalized acquisition index, coupled directly to the multicoil forward model. No equations, fitted parameters, or predictions are shown that reduce the claimed unification or efficiency to a definitional identity or self-citation chain. The central premise is an independent design choice rather than a re-derivation of prior results, with no load-bearing self-citations, ansatzes smuggled via citation, or uniqueness theorems invoked from the authors' own prior work. The derivation chain is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, mathematical axioms, or new physical entities are enumerated. The framework itself is the proposed contribution.

pith-pipeline@v0.9.1-grok · 5828 in / 1161 out tokens · 24781 ms · 2026-06-28T20:07:02.143640+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 38 canonical work pages

  1. [1]

    GRASP-Pro: imProving GRASP DCE-MRI through self-calibrating subspace-modeling and contrast phase automation.Magnetic Resonance in Medicine, 83(1):94–108, 2020

    Li Feng, Qiuting Wen, Chenchan Huang, Angela Tong, Fang Liu, and Hersh Chandarana. GRASP-Pro: imProving GRASP DCE-MRI through self-calibrating subspace-modeling and contrast phase automation.Magnetic Resonance in Medicine, 83(1):94–108, 2020. ISSN 1522-2594. doi: 10.1002/mrm.27903

    work page doi:10.1002/mrm.27903 2020

  2. [2]

    Sodickson, and Ricardo Otazo

    Li Feng, Leon Axel, Hersh Chandarana, Kai Tobias Block, Daniel K. Sodickson, and Ricardo Otazo. XD-GRASP: Golden-angle radial MRI with reconstruction of extra motion-state dimensions using compressed sensing.Magnetic Resonance in Medicine, 75(2):775–788,

  3. [3]

    doi: 10.1002/mrm.25665

    ISSN 1522-2594. doi: 10.1002/mrm.25665

    work page doi:10.1002/mrm.25665

  4. [4]

    Sodickson, and Ricardo Otazo

    Li Feng, Robert Grimm, Kai Tobias Block, Hersh Chandarana, Sungheon Kim, Jian Xu, Leon Axel, Daniel K. Sodickson, and Ricardo Otazo. Golden-angle radial sparse parallel MRI: Combination of compressed sensing, parallel imaging, and golden-angle radial sampling for fast and flexible dynamic volumetric MRI.Magnetic Resonance in Medicine, 72(3): 707–717, 2014...

    work page doi:10.1002/mrm.24980 2014

  5. [5]

    Accelerated Dynamic MRI Exploiting Sparsity and Low-Rank Structure: k-t SLR.IEEE Transactions on Medical Imaging, 30(5):1042–1054, May 2011

    Sajan Goud Lingala, Yue Hu, Edward DiBella, and Mathews Jacob. Accelerated Dynamic MRI Exploiting Sparsity and Low-Rank Structure: k-t SLR.IEEE Transactions on Medical Imaging, 30(5):1042–1054, May 2011. ISSN 1558-254X. doi: 10.1109/TMI.2010.2100850

    work page doi:10.1109/tmi.2010.2100850 2011

  6. [6]

    Sodickson

    Ricardo Otazo, Emmanuel Candès, and Daniel K. Sodickson. Low-rank plus sparse matrix decomposition for accelerated dynamic MRI with separation of background and dynamic components.Magnetic Resonance in Medicine, 73(3):1125–1136, 2015. ISSN 1522-2594. doi: 10.1002/mrm.25240

    work page doi:10.1002/mrm.25240 2015

  7. [7]

    Tolga Çukur, Salman U. H. Dar, Valiyeh A. Nezhad, Yohan Jun, Tae Hyung Kim, Shohei Fujita, and Berkin Bilgic. A tutorial on MRI reconstruction: From modern methods to clinical implications.IEEE Transactions on Biomedical Engineering, 73(5):1900–1920, 2026. doi: 10.1109/TBME.2025.3617575

    work page doi:10.1109/tbme.2025.3617575 1900

  8. [8]

    Data- and Physics-Driven Deep Learning Based Reconstruction for Fast MRI: Fundamentals and Methodologies.IEEE Reviews in Biomedical Engineering, 18:152–171,

    Jiahao Huang, Yinzhe Wu, Fanwen Wang, Yingying Fang, Yang Nan, Cagan Alkan, Daniel Abraham, Congyu Liao, Lei Xu, Zhifan Gao, Weiwen Wu, Lei Zhu, Zhaolin Chen, Peter Lally, Neal Bangerter, Kawin Setsompop, Yike Guo, Daniel Rueckert, Ge Wang, and Guang Yang. Data- and Physics-Driven Deep Learning Based Reconstruction for Fast MRI: Fundamentals and Methodolo...

  9. [9]

    doi: 10.1109/RBME.2024.3485022

    ISSN 1941-1189. doi: 10.1109/RBME.2024.3485022. 22

    work page doi:10.1109/rbme.2024.3485022 1941

  10. [10]

    Aggarwal, Merry P

    Hemant K. Aggarwal, Merry P. Mani, and Mathews Jacob. MoDL: Model-Based Deep Learning Architecture for Inverse Problems.IEEE Transactions on Medical Imaging, 38(2): 394–405, February 2019. ISSN 1558-254X. doi: 10.1109/TMI.2018.2865356

    work page doi:10.1109/tmi.2018.2865356 2019

  11. [11]

    Hajnal, Anthony N

    Jo Schlemper, Jose Caballero, Joseph V. Hajnal, Anthony N. Price, and Daniel Rueckert. A Deep cascade of convolutional neural networks for dynamic MR image reconstruction.IEEE Transactions on Medical Imaging, 37(2):491–503, 2018. doi: 10.1109/TMI.2017.2760978

    work page doi:10.1109/tmi.2017.2760978 2018

  12. [12]

    Learned low-rank priors in dynamic mr imaging.IEEE Transactions on Medical Imaging, 40(12):3698–3710, 2021

    Ziwen Ke, Wenqi Huang, Zhuo-Xu Cui, Jing Cheng, Sen Jia, Haifeng Wang, Xin Liu, Hairong Zheng, Leslie Ying, Yanjie Zhu, and Dong Liang. Learned low-rank priors in dynamic mr imaging.IEEE Transactions on Medical Imaging, 40(12):3698–3710, 2021

    2021

  13. [13]

    Deep separable spatiotemporal learning for fast dynamic cardiac MRI.IEEE Transactions on Biomedical Engineering, 72(12):3642–3654,

    Zi Wang, Min Xiao, Yirong Zhou, Chengyan Wang, Naiming Wu, Yi Li, Yiwen Gong, Shufu Chang, Yinyin Chen, Liuhong Zhu, Jianjun Zhou, Congbo Cai, He Wang, Xianwang Jiang, Di Guo, Guang Yang, and Xiaobo Qu. Deep separable spatiotemporal learning for fast dynamic cardiac MRI.IEEE Transactions on Biomedical Engineering, 72(12):3642–3654,

  14. [14]

    doi: 10.1109/TBME.2025.3574090

    work page doi:10.1109/tbme.2025.3574090 2025

  15. [15]

    One for multiple: Physics-informed synthetic data boosts generalizable deep learning for fast MRI reconstruction.Medical Image Analysis, 103:103616, 2025

    Zi Wang, Xiaotong Yu, Chengyan Wang, Weibo Chen, Jiazheng Wang, Ying-Hua Chu, Hongwei Sun, Rushuai Li, Peiyong Li, Fan Yang, Haiwei Han, Taishan Kang, Jianzhong Lin, Chen Yang, Shufu Chang, Zhang Shi, Sha Hua, Yan Li, Juan Hu, Liuhong Zhu, Jianjun Zhou, Meijing Lin, Jiefeng Guo, Congbo Cai, Zhong Chen, Di Guo, Guang Yang, and Xiaobo Qu. One for multiple: ...

    work page doi:10.1016/j.media.2025.103616 2025

  16. [16]

    Self-supervised learning of physics-guided reconstruction neural networks without fully sampled reference data.Magnetic Resonance in Medicine, 84 (6):3172–3191, 2020

    Burhaneddin Yaman, Seyed Amir Hossein Hosseini, Steen Moeller, Jutta Ellermann, Kâmil Uğurbil, and Mehmet Akçakaya. Self-supervised learning of physics-guided reconstruction neural networks without fully sampled reference data.Magnetic Resonance in Medicine, 84 (6):3172–3191, 2020. ISSN 1522-2594. doi: 10.1002/mrm.28378

    work page doi:10.1002/mrm.28378 2020

  17. [17]

    Zero-shot self-supervised learning for MRI reconstruction

    Burhaneddin Yaman, Seyed Amir Hossein Hosseini, and Mehmet Akçakaya. Zero-shot self-supervised learning for MRI reconstruction. InInternational Conference on Learning Representations, 2022. URLhttps://openreview.net/forum?id=085y6YPaYjP

    2022

  18. [18]

    Duncan, and Michal Sofka

    Bo Zhou, Jo Schlemper, Neel Dey, Seyed Sadegh Mohseni Salehi, Kevin Sheth, Chi Liu, James S. Duncan, and Michal Sofka. Dual-domain self-supervised learning for accelerated non-Cartesian MRI reconstruction.Medical Image Analysis, 81:102538, October 2022. ISSN 1361-8415. doi: 10.1016/j.media.2022.102538

    work page doi:10.1016/j.media.2022.102538 2022

  19. [19]

    Time-Dependent Deep Image Prior for Dynamic MRI.IEEE Transactions on Medical Imaging, 40(12):3337–3348, December 2021

    Jaejun Yoo, Kyong Hwan Jin, Harshit Gupta, Jérôme Yerly, Matthias Stuber, and Michael Unser. Time-Dependent Deep Image Prior for Dynamic MRI.IEEE Transactions on Medical Imaging, 40(12):3337–3348, December 2021. ISSN 1558-254X. doi: 10.1109/TMI. 2021.3084288

    work page doi:10.1109/tmi 2021

  20. [20]

    Spatiotemporal Implicit Neural Representation for Unsupervised Dynamic MRI Reconstruction.IEEE Transactions on Medical Imaging, 44(5):2143–2156, May 2025

    Jie Feng, Ruimin Feng, Qing Wu, Xin Shen, Lixuan Chen, Xin Li, Li Feng, Jingjia Chen, Zhiyong Zhang, Chunlei Liu, Yuyao Zhang, and Hongjiang Wei. Spatiotemporal Implicit Neural Representation for Unsupervised Dynamic MRI Reconstruction.IEEE Transactions on Medical Imaging, 44(5):2143–2156, May 2025. ISSN 1558-254X. doi: 10.1109/TMI.2025.3526452

    work page doi:10.1109/tmi.2025.3526452 2025

  21. [21]

    Kunz, Stefan Ruschke, and Reinhard Heckel

    Johannes F. Kunz, Stefan Ruschke, and Reinhard Heckel. Implicit Neural Networks With Fourier-Feature Inputs for Free-Breathing Cardiac MRI Reconstruction.IEEE Transactions on Computational Imaging, 10:1280–1289, 2024. ISSN 2333-9403. doi: 10.1109/TCI.2024. 3452008. 23

    work page doi:10.1109/tci.2024 2024

  22. [22]

    Jie Feng, Rui Luo, Tian Zeng, Xin Shen, Haikun Qi, Yuyao Zhang, Dong Liang, and Hongjiang Wei. Zero-shot Implicit Neural Manifold Representation (INMR) for Ultra-high Temporal Resolution Dynamic MRI.Proceedings of the AAAI Conference on Artificial Intelligence, 40(5):3930–3938, March 2026. ISSN 2374-3468. doi: 10.1609/aaai.v40i5.37395

    work page doi:10.1609/aaai.v40i5.37395 2026

  23. [23]

    Dynamic-Aware Spatio-Temporal Representation Learning for Dynamic MRI Reconstruction

    Dayoung Baik and Jaejun Yoo. Dynamic-Aware Spatio-Temporal Representation Learning for Dynamic MRI Reconstruction. In James C. Gee, Daniel C. Alexander, Jaesung Hong, Juan Eugenio Iglesias, Carole H. Sudre, Archana Venkataraman, Polina Golland, Jong Hyo Kim, and Jinah Park, editors,Medical Image Computing and Computer Assisted Intervention – MICCAI 2025, ...

    work page doi:10.1007/978-3-032-04965-0_17 2025

  24. [24]

    Schnabel, Kerstin Hammernik, Thomas Kuestner, and Daniel Rueckert

    Wenqi Huang, Veronika Spieker, Siying Xu, Gastao Cruz, Claudia Prieto, Julia A. Schnabel, Kerstin Hammernik, Thomas Kuestner, and Daniel Rueckert. Subspace Implicit Neural Representations for Real-Time Cardiac Cine MR Imaging. In Ipek Oguz, Shaoting Zhang, and Dimitris N. Metaxas, editors,Information Processing in Medical Imaging, pages 168– 183, Cham, 20...

    2026

  25. [25]

    IMJENSE: Scan-Specific Implicit Representation for Joint Coil Sensitivity and Image Estimation in Parallel MRI.IEEE Transactions on Medical Imaging, 43(4):1539–1553, April

    Ruimin Feng, Qing Wu, Jie Feng, Huajun She, Chunlei Liu, Yuyao Zhang, and Hongjiang Wei. IMJENSE: Scan-Specific Implicit Representation for Joint Coil Sensitivity and Image Estimation in Parallel MRI.IEEE Transactions on Medical Imaging, 43(4):1539–1553, April

  26. [26]

    doi: 10.1109/TMI.2023.3342156

    ISSN 1558-254X. doi: 10.1109/TMI.2023.3342156

    work page doi:10.1109/tmi.2023.3342156 2023

  27. [27]

    Xin Shen, Jie Feng, Zhenghao Li, Qing Zou, Yuyao Zhang, and Hongjiang Wei. Free- breathing dynamic MRI reconstruction via joint time-dependent coil sensitivity estimation using implicit neural representation.Medical Image Analysis, 108:103847, February 2026. ISSN 1361-8415. doi: 10.1016/j.media.2025.103847

    work page doi:10.1016/j.media.2025.103847 2026

  28. [28]

    Joint MAPLE: Accel- erated joint T1 and T2* mapping with scan-specific self-supervised networks.Magnetic Resonance in Medicine, 91(6):2294–2309, 2024

    Amir Heydari, Abbas Ahmadi, Tae Hyung Kim, and Berkin Bilgic. Joint MAPLE: Accel- erated joint T1 and T2* mapping with scan-specific self-supervised networks.Magnetic Resonance in Medicine, 91(6):2294–2309, 2024. ISSN 1522-2594. doi: 10.1002/mrm.29989

    work page doi:10.1002/mrm.29989 2024

  29. [29]

    Physics-guided self-supervised implicit neural representation for acceleratedT1ρmapping.IEEE Transactions on Biomedical Engineering, 73(5):1961–1974, May 2026

    Yuanyuan Liu, Jinwen Xie, Jianhao Wu, Zhuo-Xu Cui, Qingyong Zhu, Jing Cheng, Haifeng Wang, Zhen Song, Dong Liang, and Yanjie Zhu. Physics-guided self-supervised implicit neural representation for acceleratedT1ρmapping.IEEE Transactions on Biomedical Engineering, 73(5):1961–1974, May 2026. ISSN 0018-9294, 1558-2531. doi: 10.1109/TBME. 2025.3618476

    work page doi:10.1109/tbme 1961

  30. [30]

    Coordinate-based neural representation enabling zero-shot learning for fast 3D multiparametric quantitative MRI.Medical Image Analysis, 102:103530, May 2025

    Guoyan Lao, Ruimin Feng, Haikun Qi, Zhenfeng Lv, Qiangqiang Liu, Chunlei Liu, Yuyao Zhang, and Hongjiang Wei. Coordinate-based neural representation enabling zero-shot learning for fast 3D multiparametric quantitative MRI.Medical Image Analysis, 102:103530, May 2025. ISSN 1361-8415. doi: 10.1016/j.media.2025.103530

    work page doi:10.1016/j.media.2025.103530 2025

  31. [31]

    Rapid spatio-temporal MR fingerprinting using physics-informed implicit neural representation.Medical Image Analysis, 109:103935, March 2026

    Chaoguang Gong, Lixian Zou, Peng Li, Xingyang Wu, Yangzi Qiao, Zhanqi Hu, Xiaoyan Wang, Yihang Zhou, Kai Wang, Yue Hu, and Haifeng Wang. Rapid spatio-temporal MR fingerprinting using physics-informed implicit neural representation.Medical Image Analysis, 109:103935, March 2026. ISSN 1361-8415. doi: 10.1016/j.media.2026.103935

    work page doi:10.1016/j.media.2026.103935 2026

  32. [32]

    Pirkl, Ana Beatriz Solana, Hannah Eichhorn, Veronika Spieker, Wenqi Huang, Tim Sprenger, Marion I

    Natascha Niessen, Carolin M. Pirkl, Ana Beatriz Solana, Hannah Eichhorn, Veronika Spieker, Wenqi Huang, Tim Sprenger, Marion I. Menzel, and Julia A. Schnabel. INR Meets Multi- contrast MRI Reconstruction. In Lina Felsner, Thomas Küstner, Andreas Maier, Chen Qin, Seyed-Ahmad Ahmadi, Anees Kazi, and Xiaoling Hu, editors,Reconstruction and Imaging Motion Est...

    work page doi:10.1007/978-3-032-06103-4_3 2026

  33. [33]

    Self-supervised Deep Un- rolled Model with Implicit Neural Representation Regularization for Accelerating MRI Reconstruction, November 2025

    Jingran Xu, Yuanyuan Liu, Yuanbiao Yang, Zhuo-Xu Cui, Jing Cheng, Qingyong Zhu, Nannan Zhang, Yihang Zhou, Dong Liang, and Yanjie Zhu. Self-supervised Deep Un- rolled Model with Implicit Neural Representation Regularization for Accelerating MRI Reconstruction, November 2025. preprint arXiv:2510.06611

    work page arXiv 2025

  34. [34]

    Highly accelerated MRI via implicit neural representation guided posterior sampling of diffusion models.Medical Image Analysis, 100:103398, February 2025

    Jiayue Chu, Chenhe Du, Xiyue Lin, Xiaoqun Zhang, Lihui Wang, Yuyao Zhang, and Hongjiang Wei. Highly accelerated MRI via implicit neural representation guided posterior sampling of diffusion models.Medical Image Analysis, 100:103398, February 2025. ISSN 1361-8415. doi: 10.1016/j.media.2024.103398

    work page doi:10.1016/j.media.2024.103398 2025

  35. [35]

    Neural Implicit Dictionary learning via Mixture-of-Expert training

    Peihao Wang, Zhiwen Fan, Tianlong Chen, and Zhangyang Wang. Neural Implicit Dictionary learning via Mixture-of-Expert training. InProceedings of the 39th International Conference on Machine Learning, volume 162 ofProceedings of Machine Learning Research, pages 22613–22624. PMLR, 2022. URLhttps://proceedings.mlr.press/v162/wang22d.html

    2022

  36. [36]

    Implicit Neural Repre- sentations with Levels-of-Experts.Advances in Neural Information Processing Systems, 35:2564–2576, December 2022

    Zekun Hao, Arun Mallya, Serge Belongie, and Ming-Yu Liu. Implicit Neural Repre- sentations with Levels-of-Experts.Advances in Neural Information Processing Systems, 35:2564–2576, December 2022. URLhttps://proceedings.neurips.cc/paper_files/ paper/2022/hash/1165af8b913fb836c6280b42d6e0084f-Abstract-Conference.html

    2022

  37. [37]

    MoEC: Mixture of Experts Implicit Neural Compression, December

    Jianchen Zhao, Cheng-Ching Tseng, Ming Lu, Ruichuan An, Xiaobao Wei, He Sun, and Shanghang Zhang. MoEC: Mixture of Experts Implicit Neural Compression, December

  38. [38]

    preprint arXiv:2312.01361

    work page arXiv

  39. [39]

    Hewa Koneputugodage, Sameera Ramasinghe, and Stephen Gould

    Yizhak Ben-Shabat, Chamin P. Hewa Koneputugodage, Sameera Ramasinghe, and Stephen Gould. Neural Experts: Mixture of Experts for Implicit Neural Representations. InAdvances in Neural Information Processing Systems, volume 37, 2024. URLhttps://openreview. net/forum?id=wWguwYhpAY

    2024

  40. [40]

    Deepsd: Automatic deep skinning and pose space deformation for 3d garment animation

    Christian Reiser, Songyou Peng, Yiyi Liao, and Andreas Geiger. KiloNeRF: Speeding up Neural Radiance Fields with Thousands of Tiny MLPs. In2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 14315–14325, October 2021. doi: 10.1109/ ICCV48922.2021.01407

    work page arXiv 2021

  41. [41]

    Neural Space-Time Modeling for Motion- Corrected MR Reconstruction

    Aizada Nurdinova, Wenqi Huang, Daniel Raz Abraham, Jaehyeok Bae, Yimeng Lin, Kawin Setsompop, and Brian Andrew Hargreaves. Neural Space-Time Modeling for Motion- Corrected MR Reconstruction. In Lina Felsner, Thomas Küstner, Andreas Maier, Chen Qin, Seyed-Ahmad Ahmadi, Anees Kazi, and Xiaoling Hu, editors,Reconstruction and Imaging Motion Estimation, and G...

    work page doi:10.1007/978-3-032-06103-4_12 2026

  42. [42]

    Xuanyu Tian, Lixuan Chen, Qing Wu, Xiao Wang, Jie Feng, Yuyao Zhang, and Hongjiang Wei. Unsupervised Motion-Compensated Decomposition for Cardiac MRI Reconstruction via Neural Representation.Proceedings of the AAAI Conference on Artificial Intelligence, 40(11):9529–9537, March 2026. ISSN 2374-3468. doi: 10.1609/aaai.v40i11.37914

    work page doi:10.1609/aaai.v40i11.37914 2026

  43. [43]

    Joint image reconstruction and sensitivity estimation in SENSE (JSENSE).Magnetic Resonance in Medicine, 57(6):1196–1202, 2007

    Leslie Ying and Jinhua Sheng. Joint image reconstruction and sensitivity estimation in SENSE (JSENSE).Magnetic Resonance in Medicine, 57(6):1196–1202, 2007. ISSN 1522-2594. doi: 10.1002/mrm.21245

    work page doi:10.1002/mrm.21245 2007

  44. [44]

    Image reconstruction by regularized nonlinear inversion—Joint estimation of coil sensitivities and image content

    Martin Uecker, Thorsten Hohage, Kai Tobias Block, and Jens Frahm. Image reconstruction by regularized nonlinear inversion—Joint estimation of coil sensitivities and image content. Magnetic Resonance in Medicine, 60(3):674–682, 2008. ISSN 1522-2594. doi: 10.1002/mrm. 21691. 25

    work page doi:10.1002/mrm 2008

  45. [45]

    Murphy, Patrick Virtue, Michael Elad, John M

    Martin Uecker, Peng Lai, Mark J. Murphy, Patrick Virtue, Michael Elad, John M. Pauly, Shreyas S. Vasanawala, and Michael Lustig. ESPIRiT—an eigenvalue approach to autocali- brating parallel MRI: Where SENSE meets GRAPPA.Magnetic Resonance in Medicine, 71 (3):990–1001, 2014. ISSN 1522-2594. doi: 10.1002/mrm.24751

    work page doi:10.1002/mrm.24751 2014

  46. [46]

    Sigmoid-weighted linear units for neural network function approximation in reinforcement learning.Neural Networks, 107:3–11, November 2018

    Stefan Elfwing, Eiji Uchibe, and Kenji Doya. Sigmoid-weighted linear units for neural network function approximation in reinforcement learning.Neural Networks, 107:3–11, November 2018. ISSN 0893-6080. doi: 10.1016/j.neunet.2017.12.012

    work page doi:10.1016/j.neunet.2017.12.012 2018

  47. [47]

    CMRxRecon: A publicly available k-space dataset and benchmark to advance deep learning for cardiac MRI.Scientific Data, 11(1):687, June 2024

    Chengyan Wang, Jun Lyu, Shuo Wang, Chen Qin, Kunyuan Guo, Xinyu Zhang, Xiaotong Yu, Yan Li, Fanwen Wang, Jianhua Jin, Zhang Shi, Ziqiang Xu, Yapeng Tian, Sha Hua, Zhensen Chen, Meng Liu, Mengting Sun, Xutong Kuang, Kang Wang, Haoran Wang, Hao Li, Yinghua Chu, Guang Yang, Wenjia Bai, Xiahai Zhuang, He Wang, Jing Qin, and Xiaobo Qu. CMRxRecon: A publicly av...

    work page doi:10.1038/s41597-024-03525-4 2024

  48. [48]

    OCMR (v1.0)–Open-Access Multi-Coil k- Space Dataset for Cardiovascular Magnetic Resonance Imaging, August 2020

    Chong Chen, Yingmin Liu, Philip Schniter, Matthew Tong, Karolina Zareba, Orlando Simonetti, Lee Potter, and Rizwan Ahmad. OCMR (v1.0)–Open-Access Multi-Coil k- Space Dataset for Cardiovascular Magnetic Resonance Imaging, August 2020. preprint arXiv:2008.03410

    work page arXiv 2020

  49. [49]

    OCMR: Open-access multi-coil k-space dataset for cardiovascular magnetic reso- nance imaging

    OCMR. OCMR: Open-access multi-coil k-space dataset for cardiovascular magnetic reso- nance imaging. URLhttps://www.ocmr.info/

  50. [50]

    CMRxRecon2024: A multimodality, multiview k-space dataset boosting universal machine learning for accelerated cardiac MRI.Radiology: Artificial Intelligence, 7(2):e240443, 2025

    Zi Wang, Fanwen Wang, Chen Qin, Jun Lyu, Cheng Ouyang, Shuo Wang, Yan Li, Mengyao Yu, Haoyu Zhang, Kunyuan Guo, Zhang Shi, Qirong Li, Ziqiang Xu, Yajing Zhang, Hao Li, Sha Hua, Binghua Chen, Longyu Sun, Mengting Sun, Qing Li, Ying-Hua Chu, Wenjia Bai, Jing Qin, Xiahai Zhuang, Claudia Prieto, Alistair Young, Michael Markl, He Wang, Lian-Ming Wu, Guang Yang...

    work page doi:10.1148/ryai.240443 2025